Previous workflow engines have typically used definitions of workflows with tasks having pre-assigned resources or resources computed by earlier tasks in the workflow. Also, previous workflow engines have typically used if-then rules and conditions to specify and control execution of tasks in the workflow. In contrast, the methods described herein use constraint programming techniques. Information about a workflow is provided, comprising a plurality of tasks, and for at least some of those tasks, resource allocation requirements. Using this workflow information together with policy information and information about resource characteristics, a constraint optimization problem is specified. This problem is solved using a constraint programming solver and the resulting information about resources allocated to tasks is stored. In this way, resources may be allocated to tasks in a dynamic manner, during execution of a workflow if required.